A Hybrid Approach for Information Extraction and Expert Action Recommendation using Fine-Tuned Base Models, Large Language Models and Knowledge Graphs

ePubs

The open archive for STFC research publications

Persistent URL	http://purl.org/net/epubs/work/64997591
Record Status	Checked
Record Id	64997591
Title	A Hybrid Approach for Information Extraction and Expert Action Recommendation using Fine-Tuned Base Models, Large Language Models and Knowledge Graphs
Contributors	Dr RE Firth (STFC Daresbury Lab.) (Pr.Au.), Mr J Palmer, Ms RG George (STFC Daresbury Lab.)
Abstract	The Conveyancing process relies on legal specialists (conveyancers) requesting and exchanging information, often in an unstructured way, i.e. emails, attachments, in order to satisfy conditions for a property sale. We present a novel toolchain that uses a fine-tuned BERT-based transformer to segment and classify subsequences within emails as individual enquiries, before passing these to a keyphrase extraction model or a Large Language Model (LLM), which resolves against an expert-generated Knowledge Graph of responses and actions. These responses and actions are used to support the decisions of the Conveyancer, increasing throughput and reducing error
Organisation	STFC , HC
Keywords	natural language processing , AI , professional services , knowledge graphs , LLM , classification
Funding Information
Related Research Object(s):
Licence Information:
Language	English (EN)

Type	Details	URI(s)	Local file(s)	Year
Presentation	Presented at Advances in Data Science and Artificial Intelligence 2024, Manchester, UK, 3-4 Jun 2024.		Advances in DS an…Poster Landscape.pdf	2024